uipc_usrreq.c revision 1.147
1 /* $NetBSD: uipc_usrreq.c,v 1.147 2013/10/25 19:55:22 martin Exp $ */ 2 3 /*- 4 * Copyright (c) 1998, 2000, 2004, 2008, 2009 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility, 9 * NASA Ames Research Center, and by Andrew Doran. 10 * 11 * Redistribution and use in source and binary forms, with or without 12 * modification, are permitted provided that the following conditions 13 * are met: 14 * 1. Redistributions of source code must retain the above copyright 15 * notice, this list of conditions and the following disclaimer. 16 * 2. Redistributions in binary form must reproduce the above copyright 17 * notice, this list of conditions and the following disclaimer in the 18 * documentation and/or other materials provided with the distribution. 19 * 20 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 21 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 22 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 23 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 24 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 25 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 26 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 27 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 28 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 29 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 30 * POSSIBILITY OF SUCH DAMAGE. 31 */ 32 33 /* 34 * Copyright (c) 1982, 1986, 1989, 1991, 1993 35 * The Regents of the University of California. All rights reserved. 36 * 37 * Redistribution and use in source and binary forms, with or without 38 * modification, are permitted provided that the following conditions 39 * are met: 40 * 1. Redistributions of source code must retain the above copyright 41 * notice, this list of conditions and the following disclaimer. 42 * 2. Redistributions in binary form must reproduce the above copyright 43 * notice, this list of conditions and the following disclaimer in the 44 * documentation and/or other materials provided with the distribution. 45 * 3. Neither the name of the University nor the names of its contributors 46 * may be used to endorse or promote products derived from this software 47 * without specific prior written permission. 48 * 49 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 50 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 51 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 52 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 53 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 54 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 55 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 56 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 57 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 58 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 59 * SUCH DAMAGE. 60 * 61 * @(#)uipc_usrreq.c 8.9 (Berkeley) 5/14/95 62 */ 63 64 /* 65 * Copyright (c) 1997 Christopher G. Demetriou. All rights reserved. 66 * 67 * Redistribution and use in source and binary forms, with or without 68 * modification, are permitted provided that the following conditions 69 * are met: 70 * 1. Redistributions of source code must retain the above copyright 71 * notice, this list of conditions and the following disclaimer. 72 * 2. Redistributions in binary form must reproduce the above copyright 73 * notice, this list of conditions and the following disclaimer in the 74 * documentation and/or other materials provided with the distribution. 75 * 3. All advertising materials mentioning features or use of this software 76 * must display the following acknowledgement: 77 * This product includes software developed by the University of 78 * California, Berkeley and its contributors. 79 * 4. Neither the name of the University nor the names of its contributors 80 * may be used to endorse or promote products derived from this software 81 * without specific prior written permission. 82 * 83 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 84 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 85 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 86 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 87 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 88 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 89 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 90 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 91 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 92 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 93 * SUCH DAMAGE. 94 * 95 * @(#)uipc_usrreq.c 8.9 (Berkeley) 5/14/95 96 */ 97 98 #include <sys/cdefs.h> 99 __KERNEL_RCSID(0, "$NetBSD: uipc_usrreq.c,v 1.147 2013/10/25 19:55:22 martin Exp $"); 100 101 #include <sys/param.h> 102 #include <sys/systm.h> 103 #include <sys/proc.h> 104 #include <sys/filedesc.h> 105 #include <sys/domain.h> 106 #include <sys/protosw.h> 107 #include <sys/socket.h> 108 #include <sys/socketvar.h> 109 #include <sys/unpcb.h> 110 #include <sys/un.h> 111 #include <sys/namei.h> 112 #include <sys/vnode.h> 113 #include <sys/file.h> 114 #include <sys/stat.h> 115 #include <sys/mbuf.h> 116 #include <sys/kauth.h> 117 #include <sys/kmem.h> 118 #include <sys/atomic.h> 119 #include <sys/uidinfo.h> 120 #include <sys/kernel.h> 121 #include <sys/kthread.h> 122 123 /* 124 * Unix communications domain. 125 * 126 * TODO: 127 * RDM 128 * rethink name space problems 129 * need a proper out-of-band 130 * 131 * Notes on locking: 132 * 133 * The generic rules noted in uipc_socket2.c apply. In addition: 134 * 135 * o We have a global lock, uipc_lock. 136 * 137 * o All datagram sockets are locked by uipc_lock. 138 * 139 * o For stream socketpairs, the two endpoints are created sharing the same 140 * independent lock. Sockets presented to PRU_CONNECT2 must already have 141 * matching locks. 142 * 143 * o Stream sockets created via socket() start life with their own 144 * independent lock. 145 * 146 * o Stream connections to a named endpoint are slightly more complicated. 147 * Sockets that have called listen() have their lock pointer mutated to 148 * the global uipc_lock. When establishing a connection, the connecting 149 * socket also has its lock mutated to uipc_lock, which matches the head 150 * (listening socket). We create a new socket for accept() to return, and 151 * that also shares the head's lock. Until the connection is completely 152 * done on both ends, all three sockets are locked by uipc_lock. Once the 153 * connection is complete, the association with the head's lock is broken. 154 * The connecting socket and the socket returned from accept() have their 155 * lock pointers mutated away from uipc_lock, and back to the connecting 156 * socket's original, independent lock. The head continues to be locked 157 * by uipc_lock. 158 * 159 * o If uipc_lock is determined to be a significant source of contention, 160 * it could easily be hashed out. It is difficult to simply make it an 161 * independent lock because of visibility / garbage collection issues: 162 * if a socket has been associated with a lock at any point, that lock 163 * must remain valid until the socket is no longer visible in the system. 164 * The lock must not be freed or otherwise destroyed until any sockets 165 * that had referenced it have also been destroyed. 166 */ 167 const struct sockaddr_un sun_noname = { 168 .sun_len = offsetof(struct sockaddr_un, sun_path), 169 .sun_family = AF_LOCAL, 170 }; 171 ino_t unp_ino; /* prototype for fake inode numbers */ 172 173 struct mbuf *unp_addsockcred(struct lwp *, struct mbuf *); 174 static void unp_mark(file_t *); 175 static void unp_scan(struct mbuf *, void (*)(file_t *), int); 176 static void unp_discard_now(file_t *); 177 static void unp_discard_later(file_t *); 178 static void unp_thread(void *); 179 static void unp_thread_kick(void); 180 static kmutex_t *uipc_lock; 181 182 static kcondvar_t unp_thread_cv; 183 static lwp_t *unp_thread_lwp; 184 static SLIST_HEAD(,file) unp_thread_discard; 185 static int unp_defer; 186 187 /* 188 * Initialize Unix protocols. 189 */ 190 void 191 uipc_init(void) 192 { 193 int error; 194 195 uipc_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 196 cv_init(&unp_thread_cv, "unpgc"); 197 198 error = kthread_create(PRI_NONE, KTHREAD_MPSAFE, NULL, unp_thread, 199 NULL, &unp_thread_lwp, "unpgc"); 200 if (error != 0) 201 panic("uipc_init %d", error); 202 } 203 204 /* 205 * A connection succeeded: disassociate both endpoints from the head's 206 * lock, and make them share their own lock. There is a race here: for 207 * a very brief time one endpoint will be locked by a different lock 208 * than the other end. However, since the current thread holds the old 209 * lock (the listening socket's lock, the head) access can still only be 210 * made to one side of the connection. 211 */ 212 static void 213 unp_setpeerlocks(struct socket *so, struct socket *so2) 214 { 215 struct unpcb *unp; 216 kmutex_t *lock; 217 218 KASSERT(solocked2(so, so2)); 219 220 /* 221 * Bail out if either end of the socket is not yet fully 222 * connected or accepted. We only break the lock association 223 * with the head when the pair of sockets stand completely 224 * on their own. 225 */ 226 KASSERT(so->so_head == NULL); 227 if (so2->so_head != NULL) 228 return; 229 230 /* 231 * Drop references to old lock. A third reference (from the 232 * queue head) must be held as we still hold its lock. Bonus: 233 * we don't need to worry about garbage collecting the lock. 234 */ 235 lock = so->so_lock; 236 KASSERT(lock == uipc_lock); 237 mutex_obj_free(lock); 238 mutex_obj_free(lock); 239 240 /* 241 * Grab stream lock from the initiator and share between the two 242 * endpoints. Issue memory barrier to ensure all modifications 243 * become globally visible before the lock change. so2 is 244 * assumed not to have a stream lock, because it was created 245 * purely for the server side to accept this connection and 246 * started out life using the domain-wide lock. 247 */ 248 unp = sotounpcb(so); 249 KASSERT(unp->unp_streamlock != NULL); 250 KASSERT(sotounpcb(so2)->unp_streamlock == NULL); 251 lock = unp->unp_streamlock; 252 unp->unp_streamlock = NULL; 253 mutex_obj_hold(lock); 254 membar_exit(); 255 /* 256 * possible race if lock is not held - see comment in 257 * uipc_usrreq(PRU_ACCEPT). 258 */ 259 KASSERT(mutex_owned(lock)); 260 solockreset(so, lock); 261 solockreset(so2, lock); 262 } 263 264 /* 265 * Reset a socket's lock back to the domain-wide lock. 266 */ 267 static void 268 unp_resetlock(struct socket *so) 269 { 270 kmutex_t *olock, *nlock; 271 struct unpcb *unp; 272 273 KASSERT(solocked(so)); 274 275 olock = so->so_lock; 276 nlock = uipc_lock; 277 if (olock == nlock) 278 return; 279 unp = sotounpcb(so); 280 KASSERT(unp->unp_streamlock == NULL); 281 unp->unp_streamlock = olock; 282 mutex_obj_hold(nlock); 283 mutex_enter(nlock); 284 solockreset(so, nlock); 285 mutex_exit(olock); 286 } 287 288 static void 289 unp_free(struct unpcb *unp) 290 { 291 292 if (unp->unp_addr) 293 free(unp->unp_addr, M_SONAME); 294 if (unp->unp_streamlock != NULL) 295 mutex_obj_free(unp->unp_streamlock); 296 free(unp, M_PCB); 297 } 298 299 int 300 unp_output(struct mbuf *m, struct mbuf *control, struct unpcb *unp, 301 struct lwp *l) 302 { 303 struct socket *so2; 304 const struct sockaddr_un *sun; 305 306 so2 = unp->unp_conn->unp_socket; 307 308 KASSERT(solocked(so2)); 309 310 if (unp->unp_addr) 311 sun = unp->unp_addr; 312 else 313 sun = &sun_noname; 314 if (unp->unp_conn->unp_flags & UNP_WANTCRED) 315 control = unp_addsockcred(l, control); 316 if (sbappendaddr(&so2->so_rcv, (const struct sockaddr *)sun, m, 317 control) == 0) { 318 so2->so_rcv.sb_overflowed++; 319 unp_dispose(control); 320 m_freem(control); 321 m_freem(m); 322 return (ENOBUFS); 323 } else { 324 sorwakeup(so2); 325 return (0); 326 } 327 } 328 329 void 330 unp_setaddr(struct socket *so, struct mbuf *nam, bool peeraddr) 331 { 332 const struct sockaddr_un *sun; 333 struct unpcb *unp; 334 bool ext; 335 336 KASSERT(solocked(so)); 337 unp = sotounpcb(so); 338 ext = false; 339 340 for (;;) { 341 sun = NULL; 342 if (peeraddr) { 343 if (unp->unp_conn && unp->unp_conn->unp_addr) 344 sun = unp->unp_conn->unp_addr; 345 } else { 346 if (unp->unp_addr) 347 sun = unp->unp_addr; 348 } 349 if (sun == NULL) 350 sun = &sun_noname; 351 nam->m_len = sun->sun_len; 352 if (nam->m_len > MLEN && !ext) { 353 sounlock(so); 354 MEXTMALLOC(nam, MAXPATHLEN * 2, M_WAITOK); 355 solock(so); 356 ext = true; 357 } else { 358 KASSERT(nam->m_len <= MAXPATHLEN * 2); 359 memcpy(mtod(nam, void *), sun, (size_t)nam->m_len); 360 break; 361 } 362 } 363 } 364 365 /*ARGSUSED*/ 366 int 367 uipc_usrreq(struct socket *so, int req, struct mbuf *m, struct mbuf *nam, 368 struct mbuf *control, struct lwp *l) 369 { 370 struct unpcb *unp = sotounpcb(so); 371 struct socket *so2; 372 u_int newhiwat; 373 int error = 0; 374 375 if (req == PRU_CONTROL) 376 return (EOPNOTSUPP); 377 378 #ifdef DIAGNOSTIC 379 if (req != PRU_SEND && req != PRU_SENDOOB && control) 380 panic("uipc_usrreq: unexpected control mbuf"); 381 #endif 382 if (req != PRU_ATTACH) { 383 if (unp == NULL) { 384 error = EINVAL; 385 goto release; 386 } 387 KASSERT(solocked(so)); 388 } 389 390 switch (req) { 391 392 case PRU_ATTACH: 393 if (unp != NULL) { 394 error = EISCONN; 395 break; 396 } 397 error = unp_attach(so); 398 break; 399 400 case PRU_DETACH: 401 unp_detach(unp); 402 break; 403 404 case PRU_BIND: 405 KASSERT(l != NULL); 406 error = unp_bind(so, nam, l); 407 break; 408 409 case PRU_LISTEN: 410 /* 411 * If the socket can accept a connection, it must be 412 * locked by uipc_lock. 413 */ 414 unp_resetlock(so); 415 if (unp->unp_vnode == NULL) 416 error = EINVAL; 417 break; 418 419 case PRU_CONNECT: 420 KASSERT(l != NULL); 421 error = unp_connect(so, nam, l); 422 break; 423 424 case PRU_CONNECT2: 425 error = unp_connect2(so, (struct socket *)nam, PRU_CONNECT2); 426 break; 427 428 case PRU_DISCONNECT: 429 unp_disconnect(unp); 430 break; 431 432 case PRU_ACCEPT: 433 KASSERT(so->so_lock == uipc_lock); 434 /* 435 * Mark the initiating STREAM socket as connected *ONLY* 436 * after it's been accepted. This prevents a client from 437 * overrunning a server and receiving ECONNREFUSED. 438 */ 439 if (unp->unp_conn == NULL) { 440 /* 441 * This will use the empty socket and will not 442 * allocate. 443 */ 444 unp_setaddr(so, nam, true); 445 break; 446 } 447 so2 = unp->unp_conn->unp_socket; 448 if (so2->so_state & SS_ISCONNECTING) { 449 KASSERT(solocked2(so, so->so_head)); 450 KASSERT(solocked2(so2, so->so_head)); 451 soisconnected(so2); 452 } 453 /* 454 * If the connection is fully established, break the 455 * association with uipc_lock and give the connected 456 * pair a seperate lock to share. 457 * There is a race here: sotounpcb(so2)->unp_streamlock 458 * is not locked, so when changing so2->so_lock 459 * another thread can grab it while so->so_lock is still 460 * pointing to the (locked) uipc_lock. 461 * this should be harmless, except that this makes 462 * solocked2() and solocked() unreliable. 463 * Another problem is that unp_setaddr() expects the 464 * the socket locked. Grabing sotounpcb(so2)->unp_streamlock 465 * fixes both issues. 466 */ 467 mutex_enter(sotounpcb(so2)->unp_streamlock); 468 unp_setpeerlocks(so2, so); 469 /* 470 * Only now return peer's address, as we may need to 471 * block in order to allocate memory. 472 * 473 * XXX Minor race: connection can be broken while 474 * lock is dropped in unp_setaddr(). We will return 475 * error == 0 and sun_noname as the peer address. 476 */ 477 unp_setaddr(so, nam, true); 478 /* so_lock now points to unp_streamlock */ 479 mutex_exit(so2->so_lock); 480 break; 481 482 case PRU_SHUTDOWN: 483 socantsendmore(so); 484 unp_shutdown(unp); 485 break; 486 487 case PRU_RCVD: 488 switch (so->so_type) { 489 490 case SOCK_DGRAM: 491 panic("uipc 1"); 492 /*NOTREACHED*/ 493 494 case SOCK_SEQPACKET: /* FALLTHROUGH */ 495 case SOCK_STREAM: 496 #define rcv (&so->so_rcv) 497 #define snd (&so2->so_snd) 498 if (unp->unp_conn == 0) 499 break; 500 so2 = unp->unp_conn->unp_socket; 501 KASSERT(solocked2(so, so2)); 502 /* 503 * Adjust backpressure on sender 504 * and wakeup any waiting to write. 505 */ 506 snd->sb_mbmax += unp->unp_mbcnt - rcv->sb_mbcnt; 507 unp->unp_mbcnt = rcv->sb_mbcnt; 508 newhiwat = snd->sb_hiwat + unp->unp_cc - rcv->sb_cc; 509 (void)chgsbsize(so2->so_uidinfo, 510 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 511 unp->unp_cc = rcv->sb_cc; 512 sowwakeup(so2); 513 #undef snd 514 #undef rcv 515 break; 516 517 default: 518 panic("uipc 2"); 519 } 520 break; 521 522 case PRU_SEND: 523 /* 524 * Note: unp_internalize() rejects any control message 525 * other than SCM_RIGHTS, and only allows one. This 526 * has the side-effect of preventing a caller from 527 * forging SCM_CREDS. 528 */ 529 if (control) { 530 sounlock(so); 531 error = unp_internalize(&control); 532 solock(so); 533 if (error != 0) { 534 m_freem(control); 535 m_freem(m); 536 break; 537 } 538 } 539 switch (so->so_type) { 540 541 case SOCK_DGRAM: { 542 KASSERT(so->so_lock == uipc_lock); 543 if (nam) { 544 if ((so->so_state & SS_ISCONNECTED) != 0) 545 error = EISCONN; 546 else { 547 /* 548 * Note: once connected, the 549 * socket's lock must not be 550 * dropped until we have sent 551 * the message and disconnected. 552 * This is necessary to prevent 553 * intervening control ops, like 554 * another connection. 555 */ 556 error = unp_connect(so, nam, l); 557 } 558 } else { 559 if ((so->so_state & SS_ISCONNECTED) == 0) 560 error = ENOTCONN; 561 } 562 if (error) { 563 unp_dispose(control); 564 m_freem(control); 565 m_freem(m); 566 break; 567 } 568 KASSERT(l != NULL); 569 error = unp_output(m, control, unp, l); 570 if (nam) 571 unp_disconnect(unp); 572 break; 573 } 574 575 case SOCK_SEQPACKET: /* FALLTHROUGH */ 576 case SOCK_STREAM: 577 #define rcv (&so2->so_rcv) 578 #define snd (&so->so_snd) 579 if (unp->unp_conn == NULL) { 580 error = ENOTCONN; 581 break; 582 } 583 so2 = unp->unp_conn->unp_socket; 584 KASSERT(solocked2(so, so2)); 585 if (unp->unp_conn->unp_flags & UNP_WANTCRED) { 586 /* 587 * Credentials are passed only once on 588 * SOCK_STREAM and SOCK_SEQPACKET. 589 */ 590 unp->unp_conn->unp_flags &= ~UNP_WANTCRED; 591 control = unp_addsockcred(l, control); 592 } 593 /* 594 * Send to paired receive port, and then reduce 595 * send buffer hiwater marks to maintain backpressure. 596 * Wake up readers. 597 */ 598 if (control) { 599 if (sbappendcontrol(rcv, m, control) != 0) 600 control = NULL; 601 } else { 602 switch(so->so_type) { 603 case SOCK_SEQPACKET: 604 sbappendrecord(rcv, m); 605 break; 606 case SOCK_STREAM: 607 sbappend(rcv, m); 608 break; 609 default: 610 panic("uipc_usrreq"); 611 break; 612 } 613 } 614 snd->sb_mbmax -= 615 rcv->sb_mbcnt - unp->unp_conn->unp_mbcnt; 616 unp->unp_conn->unp_mbcnt = rcv->sb_mbcnt; 617 newhiwat = snd->sb_hiwat - 618 (rcv->sb_cc - unp->unp_conn->unp_cc); 619 (void)chgsbsize(so->so_uidinfo, 620 &snd->sb_hiwat, newhiwat, RLIM_INFINITY); 621 unp->unp_conn->unp_cc = rcv->sb_cc; 622 sorwakeup(so2); 623 #undef snd 624 #undef rcv 625 if (control != NULL) { 626 unp_dispose(control); 627 m_freem(control); 628 } 629 break; 630 631 default: 632 panic("uipc 4"); 633 } 634 break; 635 636 case PRU_ABORT: 637 (void)unp_drop(unp, ECONNABORTED); 638 639 KASSERT(so->so_head == NULL); 640 #ifdef DIAGNOSTIC 641 if (so->so_pcb == NULL) 642 panic("uipc 5: drop killed pcb"); 643 #endif 644 unp_detach(unp); 645 break; 646 647 case PRU_SENSE: 648 ((struct stat *) m)->st_blksize = so->so_snd.sb_hiwat; 649 switch (so->so_type) { 650 case SOCK_SEQPACKET: /* FALLTHROUGH */ 651 case SOCK_STREAM: 652 if (unp->unp_conn == 0) 653 break; 654 655 so2 = unp->unp_conn->unp_socket; 656 KASSERT(solocked2(so, so2)); 657 ((struct stat *) m)->st_blksize += so2->so_rcv.sb_cc; 658 break; 659 default: 660 break; 661 } 662 ((struct stat *) m)->st_dev = NODEV; 663 if (unp->unp_ino == 0) 664 unp->unp_ino = unp_ino++; 665 ((struct stat *) m)->st_atimespec = 666 ((struct stat *) m)->st_mtimespec = 667 ((struct stat *) m)->st_ctimespec = unp->unp_ctime; 668 ((struct stat *) m)->st_ino = unp->unp_ino; 669 return (0); 670 671 case PRU_RCVOOB: 672 error = EOPNOTSUPP; 673 break; 674 675 case PRU_SENDOOB: 676 m_freem(control); 677 m_freem(m); 678 error = EOPNOTSUPP; 679 break; 680 681 case PRU_SOCKADDR: 682 unp_setaddr(so, nam, false); 683 break; 684 685 case PRU_PEERADDR: 686 unp_setaddr(so, nam, true); 687 break; 688 689 default: 690 panic("piusrreq"); 691 } 692 693 release: 694 return (error); 695 } 696 697 /* 698 * Unix domain socket option processing. 699 */ 700 int 701 uipc_ctloutput(int op, struct socket *so, struct sockopt *sopt) 702 { 703 struct unpcb *unp = sotounpcb(so); 704 int optval = 0, error = 0; 705 706 KASSERT(solocked(so)); 707 708 if (sopt->sopt_level != 0) { 709 error = ENOPROTOOPT; 710 } else switch (op) { 711 712 case PRCO_SETOPT: 713 switch (sopt->sopt_name) { 714 case LOCAL_CREDS: 715 case LOCAL_CONNWAIT: 716 error = sockopt_getint(sopt, &optval); 717 if (error) 718 break; 719 switch (sopt->sopt_name) { 720 #define OPTSET(bit) \ 721 if (optval) \ 722 unp->unp_flags |= (bit); \ 723 else \ 724 unp->unp_flags &= ~(bit); 725 726 case LOCAL_CREDS: 727 OPTSET(UNP_WANTCRED); 728 break; 729 case LOCAL_CONNWAIT: 730 OPTSET(UNP_CONNWAIT); 731 break; 732 } 733 break; 734 #undef OPTSET 735 736 default: 737 error = ENOPROTOOPT; 738 break; 739 } 740 break; 741 742 case PRCO_GETOPT: 743 sounlock(so); 744 switch (sopt->sopt_name) { 745 case LOCAL_PEEREID: 746 if (unp->unp_flags & UNP_EIDSVALID) { 747 error = sockopt_set(sopt, 748 &unp->unp_connid, sizeof(unp->unp_connid)); 749 } else { 750 error = EINVAL; 751 } 752 break; 753 case LOCAL_CREDS: 754 #define OPTBIT(bit) (unp->unp_flags & (bit) ? 1 : 0) 755 756 optval = OPTBIT(UNP_WANTCRED); 757 error = sockopt_setint(sopt, optval); 758 break; 759 #undef OPTBIT 760 761 default: 762 error = ENOPROTOOPT; 763 break; 764 } 765 solock(so); 766 break; 767 } 768 return (error); 769 } 770 771 /* 772 * Both send and receive buffers are allocated PIPSIZ bytes of buffering 773 * for stream sockets, although the total for sender and receiver is 774 * actually only PIPSIZ. 775 * Datagram sockets really use the sendspace as the maximum datagram size, 776 * and don't really want to reserve the sendspace. Their recvspace should 777 * be large enough for at least one max-size datagram plus address. 778 */ 779 #define PIPSIZ 4096 780 u_long unpst_sendspace = PIPSIZ; 781 u_long unpst_recvspace = PIPSIZ; 782 u_long unpdg_sendspace = 2*1024; /* really max datagram size */ 783 u_long unpdg_recvspace = 4*1024; 784 785 u_int unp_rights; /* files in flight */ 786 u_int unp_rights_ratio = 2; /* limit, fraction of maxfiles */ 787 788 int 789 unp_attach(struct socket *so) 790 { 791 struct unpcb *unp; 792 int error; 793 794 switch (so->so_type) { 795 case SOCK_SEQPACKET: /* FALLTHROUGH */ 796 case SOCK_STREAM: 797 if (so->so_lock == NULL) { 798 /* 799 * XXX Assuming that no socket locks are held, 800 * as this call may sleep. 801 */ 802 so->so_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 803 solock(so); 804 } 805 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 806 error = soreserve(so, unpst_sendspace, unpst_recvspace); 807 if (error != 0) 808 return (error); 809 } 810 break; 811 812 case SOCK_DGRAM: 813 if (so->so_lock == NULL) { 814 mutex_obj_hold(uipc_lock); 815 so->so_lock = uipc_lock; 816 solock(so); 817 } 818 if (so->so_snd.sb_hiwat == 0 || so->so_rcv.sb_hiwat == 0) { 819 error = soreserve(so, unpdg_sendspace, unpdg_recvspace); 820 if (error != 0) 821 return (error); 822 } 823 break; 824 825 default: 826 panic("unp_attach"); 827 } 828 KASSERT(solocked(so)); 829 unp = malloc(sizeof(*unp), M_PCB, M_NOWAIT); 830 if (unp == NULL) 831 return (ENOBUFS); 832 memset(unp, 0, sizeof(*unp)); 833 unp->unp_socket = so; 834 so->so_pcb = unp; 835 nanotime(&unp->unp_ctime); 836 return (0); 837 } 838 839 void 840 unp_detach(struct unpcb *unp) 841 { 842 struct socket *so; 843 vnode_t *vp; 844 845 so = unp->unp_socket; 846 847 retry: 848 if ((vp = unp->unp_vnode) != NULL) { 849 sounlock(so); 850 /* Acquire v_interlock to protect against unp_connect(). */ 851 /* XXXAD racy */ 852 mutex_enter(vp->v_interlock); 853 vp->v_socket = NULL; 854 vrelel(vp, 0); 855 solock(so); 856 unp->unp_vnode = NULL; 857 } 858 if (unp->unp_conn) 859 unp_disconnect(unp); 860 while (unp->unp_refs) { 861 KASSERT(solocked2(so, unp->unp_refs->unp_socket)); 862 if (unp_drop(unp->unp_refs, ECONNRESET)) { 863 solock(so); 864 goto retry; 865 } 866 } 867 soisdisconnected(so); 868 so->so_pcb = NULL; 869 if (unp_rights) { 870 /* 871 * Normally the receive buffer is flushed later, in sofree, 872 * but if our receive buffer holds references to files that 873 * are now garbage, we will enqueue those file references to 874 * the garbage collector and kick it into action. 875 */ 876 sorflush(so); 877 unp_free(unp); 878 unp_thread_kick(); 879 } else 880 unp_free(unp); 881 } 882 883 /* 884 * Allocate the new sockaddr. We have to allocate one 885 * extra byte so that we can ensure that the pathname 886 * is nul-terminated. Note that unlike linux, we don't 887 * include in the address length the NUL in the path 888 * component, because doing so, would exceed sizeof(sockaddr_un) 889 * for fully occupied pathnames. Linux is also inconsistent, 890 * because it does not include the NUL in the length of 891 * what it calls "abstract" unix sockets. 892 */ 893 static struct sockaddr_un * 894 makeun(struct mbuf *nam, size_t *addrlen) { 895 struct sockaddr_un *sun; 896 897 *addrlen = nam->m_len + 1; 898 sun = malloc(*addrlen, M_SONAME, M_WAITOK); 899 m_copydata(nam, 0, nam->m_len, (void *)sun); 900 *(((char *)sun) + nam->m_len) = '\0'; 901 sun->sun_len = strlen(sun->sun_path) + 902 offsetof(struct sockaddr_un, sun_path); 903 return sun; 904 } 905 906 int 907 unp_bind(struct socket *so, struct mbuf *nam, struct lwp *l) 908 { 909 struct sockaddr_un *sun; 910 struct unpcb *unp; 911 vnode_t *vp; 912 struct vattr vattr; 913 size_t addrlen; 914 int error; 915 struct pathbuf *pb; 916 struct nameidata nd; 917 proc_t *p; 918 919 unp = sotounpcb(so); 920 if (unp->unp_vnode != NULL) 921 return (EINVAL); 922 if ((unp->unp_flags & UNP_BUSY) != 0) { 923 /* 924 * EALREADY may not be strictly accurate, but since this 925 * is a major application error it's hardly a big deal. 926 */ 927 return (EALREADY); 928 } 929 unp->unp_flags |= UNP_BUSY; 930 sounlock(so); 931 932 p = l->l_proc; 933 sun = makeun(nam, &addrlen); 934 935 pb = pathbuf_create(sun->sun_path); 936 if (pb == NULL) { 937 error = ENOMEM; 938 goto bad; 939 } 940 NDINIT(&nd, CREATE, FOLLOW | LOCKPARENT | TRYEMULROOT, pb); 941 942 /* SHOULD BE ABLE TO ADOPT EXISTING AND wakeup() ALA FIFO's */ 943 if ((error = namei(&nd)) != 0) { 944 pathbuf_destroy(pb); 945 goto bad; 946 } 947 vp = nd.ni_vp; 948 if (vp != NULL) { 949 VOP_ABORTOP(nd.ni_dvp, &nd.ni_cnd); 950 if (nd.ni_dvp == vp) 951 vrele(nd.ni_dvp); 952 else 953 vput(nd.ni_dvp); 954 vrele(vp); 955 pathbuf_destroy(pb); 956 error = EADDRINUSE; 957 goto bad; 958 } 959 vattr_null(&vattr); 960 vattr.va_type = VSOCK; 961 vattr.va_mode = ACCESSPERMS & ~(p->p_cwdi->cwdi_cmask); 962 error = VOP_CREATE(nd.ni_dvp, &nd.ni_vp, &nd.ni_cnd, &vattr); 963 if (error) { 964 pathbuf_destroy(pb); 965 goto bad; 966 } 967 vp = nd.ni_vp; 968 solock(so); 969 vp->v_socket = unp->unp_socket; 970 unp->unp_vnode = vp; 971 unp->unp_addrlen = addrlen; 972 unp->unp_addr = sun; 973 unp->unp_connid.unp_pid = p->p_pid; 974 unp->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred); 975 unp->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred); 976 unp->unp_flags |= UNP_EIDSBIND; 977 VOP_UNLOCK(vp); 978 unp->unp_flags &= ~UNP_BUSY; 979 pathbuf_destroy(pb); 980 return (0); 981 982 bad: 983 free(sun, M_SONAME); 984 solock(so); 985 unp->unp_flags &= ~UNP_BUSY; 986 return (error); 987 } 988 989 int 990 unp_connect(struct socket *so, struct mbuf *nam, struct lwp *l) 991 { 992 struct sockaddr_un *sun; 993 vnode_t *vp; 994 struct socket *so2, *so3; 995 struct unpcb *unp, *unp2, *unp3; 996 size_t addrlen; 997 int error; 998 struct pathbuf *pb; 999 struct nameidata nd; 1000 1001 unp = sotounpcb(so); 1002 if ((unp->unp_flags & UNP_BUSY) != 0) { 1003 /* 1004 * EALREADY may not be strictly accurate, but since this 1005 * is a major application error it's hardly a big deal. 1006 */ 1007 return (EALREADY); 1008 } 1009 unp->unp_flags |= UNP_BUSY; 1010 sounlock(so); 1011 1012 sun = makeun(nam, &addrlen); 1013 pb = pathbuf_create(sun->sun_path); 1014 if (pb == NULL) { 1015 error = ENOMEM; 1016 goto bad2; 1017 } 1018 1019 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb); 1020 1021 if ((error = namei(&nd)) != 0) { 1022 pathbuf_destroy(pb); 1023 goto bad2; 1024 } 1025 vp = nd.ni_vp; 1026 if (vp->v_type != VSOCK) { 1027 error = ENOTSOCK; 1028 goto bad; 1029 } 1030 pathbuf_destroy(pb); 1031 if ((error = VOP_ACCESS(vp, VWRITE, l->l_cred)) != 0) 1032 goto bad; 1033 /* Acquire v_interlock to protect against unp_detach(). */ 1034 mutex_enter(vp->v_interlock); 1035 so2 = vp->v_socket; 1036 if (so2 == NULL) { 1037 mutex_exit(vp->v_interlock); 1038 error = ECONNREFUSED; 1039 goto bad; 1040 } 1041 if (so->so_type != so2->so_type) { 1042 mutex_exit(vp->v_interlock); 1043 error = EPROTOTYPE; 1044 goto bad; 1045 } 1046 solock(so); 1047 unp_resetlock(so); 1048 mutex_exit(vp->v_interlock); 1049 if ((so->so_proto->pr_flags & PR_CONNREQUIRED) != 0) { 1050 /* 1051 * This may seem somewhat fragile but is OK: if we can 1052 * see SO_ACCEPTCONN set on the endpoint, then it must 1053 * be locked by the domain-wide uipc_lock. 1054 */ 1055 KASSERT((so2->so_options & SO_ACCEPTCONN) == 0 || 1056 so2->so_lock == uipc_lock); 1057 if ((so2->so_options & SO_ACCEPTCONN) == 0 || 1058 (so3 = sonewconn(so2, false)) == NULL) { 1059 error = ECONNREFUSED; 1060 sounlock(so); 1061 goto bad; 1062 } 1063 unp2 = sotounpcb(so2); 1064 unp3 = sotounpcb(so3); 1065 if (unp2->unp_addr) { 1066 unp3->unp_addr = malloc(unp2->unp_addrlen, 1067 M_SONAME, M_WAITOK); 1068 memcpy(unp3->unp_addr, unp2->unp_addr, 1069 unp2->unp_addrlen); 1070 unp3->unp_addrlen = unp2->unp_addrlen; 1071 } 1072 unp3->unp_flags = unp2->unp_flags; 1073 unp3->unp_connid.unp_pid = l->l_proc->p_pid; 1074 unp3->unp_connid.unp_euid = kauth_cred_geteuid(l->l_cred); 1075 unp3->unp_connid.unp_egid = kauth_cred_getegid(l->l_cred); 1076 unp3->unp_flags |= UNP_EIDSVALID; 1077 if (unp2->unp_flags & UNP_EIDSBIND) { 1078 unp->unp_connid = unp2->unp_connid; 1079 unp->unp_flags |= UNP_EIDSVALID; 1080 } 1081 so2 = so3; 1082 } 1083 error = unp_connect2(so, so2, PRU_CONNECT); 1084 sounlock(so); 1085 bad: 1086 vput(vp); 1087 bad2: 1088 free(sun, M_SONAME); 1089 solock(so); 1090 unp->unp_flags &= ~UNP_BUSY; 1091 return (error); 1092 } 1093 1094 int 1095 unp_connect2(struct socket *so, struct socket *so2, int req) 1096 { 1097 struct unpcb *unp = sotounpcb(so); 1098 struct unpcb *unp2; 1099 1100 if (so2->so_type != so->so_type) 1101 return (EPROTOTYPE); 1102 1103 /* 1104 * All three sockets involved must be locked by same lock: 1105 * 1106 * local endpoint (so) 1107 * remote endpoint (so2) 1108 * queue head (so2->so_head, only if PR_CONNREQUIRED) 1109 */ 1110 KASSERT(solocked2(so, so2)); 1111 KASSERT(so->so_head == NULL); 1112 if (so2->so_head != NULL) { 1113 KASSERT(so2->so_lock == uipc_lock); 1114 KASSERT(solocked2(so2, so2->so_head)); 1115 } 1116 1117 unp2 = sotounpcb(so2); 1118 unp->unp_conn = unp2; 1119 switch (so->so_type) { 1120 1121 case SOCK_DGRAM: 1122 unp->unp_nextref = unp2->unp_refs; 1123 unp2->unp_refs = unp; 1124 soisconnected(so); 1125 break; 1126 1127 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1128 case SOCK_STREAM: 1129 unp2->unp_conn = unp; 1130 if (req == PRU_CONNECT && 1131 ((unp->unp_flags | unp2->unp_flags) & UNP_CONNWAIT)) 1132 soisconnecting(so); 1133 else 1134 soisconnected(so); 1135 soisconnected(so2); 1136 /* 1137 * If the connection is fully established, break the 1138 * association with uipc_lock and give the connected 1139 * pair a seperate lock to share. For CONNECT2, we 1140 * require that the locks already match (the sockets 1141 * are created that way). 1142 */ 1143 if (req == PRU_CONNECT) { 1144 KASSERT(so2->so_head != NULL); 1145 unp_setpeerlocks(so, so2); 1146 } 1147 break; 1148 1149 default: 1150 panic("unp_connect2"); 1151 } 1152 return (0); 1153 } 1154 1155 void 1156 unp_disconnect(struct unpcb *unp) 1157 { 1158 struct unpcb *unp2 = unp->unp_conn; 1159 struct socket *so; 1160 1161 if (unp2 == 0) 1162 return; 1163 unp->unp_conn = 0; 1164 so = unp->unp_socket; 1165 switch (so->so_type) { 1166 case SOCK_DGRAM: 1167 if (unp2->unp_refs == unp) 1168 unp2->unp_refs = unp->unp_nextref; 1169 else { 1170 unp2 = unp2->unp_refs; 1171 for (;;) { 1172 KASSERT(solocked2(so, unp2->unp_socket)); 1173 if (unp2 == 0) 1174 panic("unp_disconnect"); 1175 if (unp2->unp_nextref == unp) 1176 break; 1177 unp2 = unp2->unp_nextref; 1178 } 1179 unp2->unp_nextref = unp->unp_nextref; 1180 } 1181 unp->unp_nextref = 0; 1182 so->so_state &= ~SS_ISCONNECTED; 1183 break; 1184 1185 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1186 case SOCK_STREAM: 1187 KASSERT(solocked2(so, unp2->unp_socket)); 1188 soisdisconnected(so); 1189 unp2->unp_conn = 0; 1190 soisdisconnected(unp2->unp_socket); 1191 break; 1192 } 1193 } 1194 1195 #ifdef notdef 1196 unp_abort(struct unpcb *unp) 1197 { 1198 unp_detach(unp); 1199 } 1200 #endif 1201 1202 void 1203 unp_shutdown(struct unpcb *unp) 1204 { 1205 struct socket *so; 1206 1207 switch(unp->unp_socket->so_type) { 1208 case SOCK_SEQPACKET: /* FALLTHROUGH */ 1209 case SOCK_STREAM: 1210 if (unp->unp_conn && (so = unp->unp_conn->unp_socket)) 1211 socantrcvmore(so); 1212 break; 1213 default: 1214 break; 1215 } 1216 } 1217 1218 bool 1219 unp_drop(struct unpcb *unp, int errno) 1220 { 1221 struct socket *so = unp->unp_socket; 1222 1223 KASSERT(solocked(so)); 1224 1225 so->so_error = errno; 1226 unp_disconnect(unp); 1227 if (so->so_head) { 1228 so->so_pcb = NULL; 1229 /* sofree() drops the socket lock */ 1230 sofree(so); 1231 unp_free(unp); 1232 return true; 1233 } 1234 return false; 1235 } 1236 1237 #ifdef notdef 1238 unp_drain(void) 1239 { 1240 1241 } 1242 #endif 1243 1244 int 1245 unp_externalize(struct mbuf *rights, struct lwp *l, int flags) 1246 { 1247 struct cmsghdr * const cm = mtod(rights, struct cmsghdr *); 1248 struct proc * const p = l->l_proc; 1249 file_t **rp; 1250 int error = 0; 1251 1252 const size_t nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / 1253 sizeof(file_t *); 1254 if (nfds == 0) 1255 goto noop; 1256 1257 int * const fdp = kmem_alloc(nfds * sizeof(int), KM_SLEEP); 1258 rw_enter(&p->p_cwdi->cwdi_lock, RW_READER); 1259 1260 /* Make sure the recipient should be able to see the files.. */ 1261 rp = (file_t **)CMSG_DATA(cm); 1262 for (size_t i = 0; i < nfds; i++) { 1263 file_t * const fp = *rp++; 1264 if (fp == NULL) { 1265 error = EINVAL; 1266 goto out; 1267 } 1268 /* 1269 * If we are in a chroot'ed directory, and 1270 * someone wants to pass us a directory, make 1271 * sure it's inside the subtree we're allowed 1272 * to access. 1273 */ 1274 if (p->p_cwdi->cwdi_rdir != NULL && fp->f_type == DTYPE_VNODE) { 1275 vnode_t *vp = (vnode_t *)fp->f_data; 1276 if ((vp->v_type == VDIR) && 1277 !vn_isunder(vp, p->p_cwdi->cwdi_rdir, l)) { 1278 error = EPERM; 1279 goto out; 1280 } 1281 } 1282 } 1283 1284 restart: 1285 /* 1286 * First loop -- allocate file descriptor table slots for the 1287 * new files. 1288 */ 1289 for (size_t i = 0; i < nfds; i++) { 1290 if ((error = fd_alloc(p, 0, &fdp[i])) != 0) { 1291 /* 1292 * Back out what we've done so far. 1293 */ 1294 while (i-- > 0) { 1295 fd_abort(p, NULL, fdp[i]); 1296 } 1297 if (error == ENOSPC) { 1298 fd_tryexpand(p); 1299 error = 0; 1300 goto restart; 1301 } 1302 /* 1303 * This is the error that has historically 1304 * been returned, and some callers may 1305 * expect it. 1306 */ 1307 error = EMSGSIZE; 1308 goto out; 1309 } 1310 } 1311 1312 /* 1313 * Now that adding them has succeeded, update all of the 1314 * file passing state and affix the descriptors. 1315 */ 1316 rp = (file_t **)CMSG_DATA(cm); 1317 int *ofdp = (int *)CMSG_DATA(cm); 1318 for (size_t i = 0; i < nfds; i++) { 1319 file_t * const fp = *rp++; 1320 const int fd = fdp[i]; 1321 atomic_dec_uint(&unp_rights); 1322 fd_set_exclose(l, fd, (flags & O_CLOEXEC) != 0); 1323 fd_affix(p, fp, fd); 1324 /* 1325 * Done with this file pointer, replace it with a fd; 1326 */ 1327 *ofdp++ = fd; 1328 mutex_enter(&fp->f_lock); 1329 fp->f_msgcount--; 1330 mutex_exit(&fp->f_lock); 1331 /* 1332 * Note that fd_affix() adds a reference to the file. 1333 * The file may already have been closed by another 1334 * LWP in the process, so we must drop the reference 1335 * added by unp_internalize() with closef(). 1336 */ 1337 closef(fp); 1338 } 1339 1340 /* 1341 * Adjust length, in case of transition from large file_t 1342 * pointers to ints. 1343 */ 1344 if (sizeof(file_t *) != sizeof(int)) { 1345 cm->cmsg_len = CMSG_LEN(nfds * sizeof(int)); 1346 rights->m_len = CMSG_SPACE(nfds * sizeof(int)); 1347 } 1348 out: 1349 if (__predict_false(error != 0)) { 1350 file_t **const fpp = (file_t **)CMSG_DATA(cm); 1351 for (size_t i = 0; i < nfds; i++) 1352 unp_discard_now(fpp[i]); 1353 /* 1354 * Truncate the array so that nobody will try to interpret 1355 * what is now garbage in it. 1356 */ 1357 cm->cmsg_len = CMSG_LEN(0); 1358 rights->m_len = CMSG_SPACE(0); 1359 } 1360 rw_exit(&p->p_cwdi->cwdi_lock); 1361 kmem_free(fdp, nfds * sizeof(int)); 1362 1363 noop: 1364 /* 1365 * Don't disclose kernel memory in the alignment space. 1366 */ 1367 KASSERT(cm->cmsg_len <= rights->m_len); 1368 memset(&mtod(rights, char *)[cm->cmsg_len], 0, rights->m_len - 1369 cm->cmsg_len); 1370 return error; 1371 } 1372 1373 int 1374 unp_internalize(struct mbuf **controlp) 1375 { 1376 filedesc_t *fdescp = curlwp->l_fd; 1377 struct mbuf *control = *controlp; 1378 struct cmsghdr *newcm, *cm = mtod(control, struct cmsghdr *); 1379 file_t **rp, **files; 1380 file_t *fp; 1381 int i, fd, *fdp; 1382 int nfds, error; 1383 u_int maxmsg; 1384 1385 error = 0; 1386 newcm = NULL; 1387 1388 /* Sanity check the control message header. */ 1389 if (cm->cmsg_type != SCM_RIGHTS || cm->cmsg_level != SOL_SOCKET || 1390 cm->cmsg_len > control->m_len || 1391 cm->cmsg_len < CMSG_ALIGN(sizeof(*cm))) 1392 return (EINVAL); 1393 1394 /* 1395 * Verify that the file descriptors are valid, and acquire 1396 * a reference to each. 1397 */ 1398 nfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) / sizeof(int); 1399 fdp = (int *)CMSG_DATA(cm); 1400 maxmsg = maxfiles / unp_rights_ratio; 1401 for (i = 0; i < nfds; i++) { 1402 fd = *fdp++; 1403 if (atomic_inc_uint_nv(&unp_rights) > maxmsg) { 1404 atomic_dec_uint(&unp_rights); 1405 nfds = i; 1406 error = EAGAIN; 1407 goto out; 1408 } 1409 if ((fp = fd_getfile(fd)) == NULL 1410 || fp->f_type == DTYPE_KQUEUE) { 1411 if (fp) 1412 fd_putfile(fd); 1413 atomic_dec_uint(&unp_rights); 1414 nfds = i; 1415 error = EBADF; 1416 goto out; 1417 } 1418 } 1419 1420 /* Allocate new space and copy header into it. */ 1421 newcm = malloc(CMSG_SPACE(nfds * sizeof(file_t *)), M_MBUF, M_WAITOK); 1422 if (newcm == NULL) { 1423 error = E2BIG; 1424 goto out; 1425 } 1426 memcpy(newcm, cm, sizeof(struct cmsghdr)); 1427 files = (file_t **)CMSG_DATA(newcm); 1428 1429 /* 1430 * Transform the file descriptors into file_t pointers, in 1431 * reverse order so that if pointers are bigger than ints, the 1432 * int won't get until we're done. No need to lock, as we have 1433 * already validated the descriptors with fd_getfile(). 1434 */ 1435 fdp = (int *)CMSG_DATA(cm) + nfds; 1436 rp = files + nfds; 1437 for (i = 0; i < nfds; i++) { 1438 fp = fdescp->fd_dt->dt_ff[*--fdp]->ff_file; 1439 KASSERT(fp != NULL); 1440 mutex_enter(&fp->f_lock); 1441 *--rp = fp; 1442 fp->f_count++; 1443 fp->f_msgcount++; 1444 mutex_exit(&fp->f_lock); 1445 } 1446 1447 out: 1448 /* Release descriptor references. */ 1449 fdp = (int *)CMSG_DATA(cm); 1450 for (i = 0; i < nfds; i++) { 1451 fd_putfile(*fdp++); 1452 if (error != 0) { 1453 atomic_dec_uint(&unp_rights); 1454 } 1455 } 1456 1457 if (error == 0) { 1458 if (control->m_flags & M_EXT) { 1459 m_freem(control); 1460 *controlp = control = m_get(M_WAIT, MT_CONTROL); 1461 } 1462 MEXTADD(control, newcm, CMSG_SPACE(nfds * sizeof(file_t *)), 1463 M_MBUF, NULL, NULL); 1464 cm = newcm; 1465 /* 1466 * Adjust message & mbuf to note amount of space 1467 * actually used. 1468 */ 1469 cm->cmsg_len = CMSG_LEN(nfds * sizeof(file_t *)); 1470 control->m_len = CMSG_SPACE(nfds * sizeof(file_t *)); 1471 } 1472 1473 return error; 1474 } 1475 1476 struct mbuf * 1477 unp_addsockcred(struct lwp *l, struct mbuf *control) 1478 { 1479 struct sockcred *sc; 1480 struct mbuf *m; 1481 void *p; 1482 1483 m = sbcreatecontrol1(&p, SOCKCREDSIZE(kauth_cred_ngroups(l->l_cred)), 1484 SCM_CREDS, SOL_SOCKET, M_WAITOK); 1485 if (m == NULL) 1486 return control; 1487 1488 sc = p; 1489 sc->sc_uid = kauth_cred_getuid(l->l_cred); 1490 sc->sc_euid = kauth_cred_geteuid(l->l_cred); 1491 sc->sc_gid = kauth_cred_getgid(l->l_cred); 1492 sc->sc_egid = kauth_cred_getegid(l->l_cred); 1493 sc->sc_ngroups = kauth_cred_ngroups(l->l_cred); 1494 1495 for (int i = 0; i < sc->sc_ngroups; i++) 1496 sc->sc_groups[i] = kauth_cred_group(l->l_cred, i); 1497 1498 return m_add(control, m); 1499 } 1500 1501 /* 1502 * Do a mark-sweep GC of files in the system, to free up any which are 1503 * caught in flight to an about-to-be-closed socket. Additionally, 1504 * process deferred file closures. 1505 */ 1506 static void 1507 unp_gc(file_t *dp) 1508 { 1509 extern struct domain unixdomain; 1510 file_t *fp, *np; 1511 struct socket *so, *so1; 1512 u_int i, old, new; 1513 bool didwork; 1514 1515 KASSERT(curlwp == unp_thread_lwp); 1516 KASSERT(mutex_owned(&filelist_lock)); 1517 1518 /* 1519 * First, process deferred file closures. 1520 */ 1521 while (!SLIST_EMPTY(&unp_thread_discard)) { 1522 fp = SLIST_FIRST(&unp_thread_discard); 1523 KASSERT(fp->f_unpcount > 0); 1524 KASSERT(fp->f_count > 0); 1525 KASSERT(fp->f_msgcount > 0); 1526 KASSERT(fp->f_count >= fp->f_unpcount); 1527 KASSERT(fp->f_count >= fp->f_msgcount); 1528 KASSERT(fp->f_msgcount >= fp->f_unpcount); 1529 SLIST_REMOVE_HEAD(&unp_thread_discard, f_unplist); 1530 i = fp->f_unpcount; 1531 fp->f_unpcount = 0; 1532 mutex_exit(&filelist_lock); 1533 for (; i != 0; i--) { 1534 unp_discard_now(fp); 1535 } 1536 mutex_enter(&filelist_lock); 1537 } 1538 1539 /* 1540 * Clear mark bits. Ensure that we don't consider new files 1541 * entering the file table during this loop (they will not have 1542 * FSCAN set). 1543 */ 1544 unp_defer = 0; 1545 LIST_FOREACH(fp, &filehead, f_list) { 1546 for (old = fp->f_flag;; old = new) { 1547 new = atomic_cas_uint(&fp->f_flag, old, 1548 (old | FSCAN) & ~(FMARK|FDEFER)); 1549 if (__predict_true(old == new)) { 1550 break; 1551 } 1552 } 1553 } 1554 1555 /* 1556 * Iterate over the set of sockets, marking ones believed (based on 1557 * refcount) to be referenced from a process, and marking for rescan 1558 * sockets which are queued on a socket. Recan continues descending 1559 * and searching for sockets referenced by sockets (FDEFER), until 1560 * there are no more socket->socket references to be discovered. 1561 */ 1562 do { 1563 didwork = false; 1564 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) { 1565 KASSERT(mutex_owned(&filelist_lock)); 1566 np = LIST_NEXT(fp, f_list); 1567 mutex_enter(&fp->f_lock); 1568 if ((fp->f_flag & FDEFER) != 0) { 1569 atomic_and_uint(&fp->f_flag, ~FDEFER); 1570 unp_defer--; 1571 KASSERT(fp->f_count != 0); 1572 } else { 1573 if (fp->f_count == 0 || 1574 (fp->f_flag & FMARK) != 0 || 1575 fp->f_count == fp->f_msgcount || 1576 fp->f_unpcount != 0) { 1577 mutex_exit(&fp->f_lock); 1578 continue; 1579 } 1580 } 1581 atomic_or_uint(&fp->f_flag, FMARK); 1582 1583 if (fp->f_type != DTYPE_SOCKET || 1584 (so = fp->f_data) == NULL || 1585 so->so_proto->pr_domain != &unixdomain || 1586 (so->so_proto->pr_flags & PR_RIGHTS) == 0) { 1587 mutex_exit(&fp->f_lock); 1588 continue; 1589 } 1590 1591 /* Gain file ref, mark our position, and unlock. */ 1592 didwork = true; 1593 LIST_INSERT_AFTER(fp, dp, f_list); 1594 fp->f_count++; 1595 mutex_exit(&fp->f_lock); 1596 mutex_exit(&filelist_lock); 1597 1598 /* 1599 * Mark files referenced from sockets queued on the 1600 * accept queue as well. 1601 */ 1602 solock(so); 1603 unp_scan(so->so_rcv.sb_mb, unp_mark, 0); 1604 if ((so->so_options & SO_ACCEPTCONN) != 0) { 1605 TAILQ_FOREACH(so1, &so->so_q0, so_qe) { 1606 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1607 } 1608 TAILQ_FOREACH(so1, &so->so_q, so_qe) { 1609 unp_scan(so1->so_rcv.sb_mb, unp_mark, 0); 1610 } 1611 } 1612 sounlock(so); 1613 1614 /* Re-lock and restart from where we left off. */ 1615 closef(fp); 1616 mutex_enter(&filelist_lock); 1617 np = LIST_NEXT(dp, f_list); 1618 LIST_REMOVE(dp, f_list); 1619 } 1620 /* 1621 * Bail early if we did nothing in the loop above. Could 1622 * happen because of concurrent activity causing unp_defer 1623 * to get out of sync. 1624 */ 1625 } while (unp_defer != 0 && didwork); 1626 1627 /* 1628 * Sweep pass. 1629 * 1630 * We grab an extra reference to each of the files that are 1631 * not otherwise accessible and then free the rights that are 1632 * stored in messages on them. 1633 */ 1634 for (fp = LIST_FIRST(&filehead); fp != NULL; fp = np) { 1635 KASSERT(mutex_owned(&filelist_lock)); 1636 np = LIST_NEXT(fp, f_list); 1637 mutex_enter(&fp->f_lock); 1638 1639 /* 1640 * Ignore non-sockets. 1641 * Ignore dead sockets, or sockets with pending close. 1642 * Ignore sockets obviously referenced elsewhere. 1643 * Ignore sockets marked as referenced by our scan. 1644 * Ignore new sockets that did not exist during the scan. 1645 */ 1646 if (fp->f_type != DTYPE_SOCKET || 1647 fp->f_count == 0 || fp->f_unpcount != 0 || 1648 fp->f_count != fp->f_msgcount || 1649 (fp->f_flag & (FMARK | FSCAN)) != FSCAN) { 1650 mutex_exit(&fp->f_lock); 1651 continue; 1652 } 1653 1654 /* Gain file ref, mark our position, and unlock. */ 1655 LIST_INSERT_AFTER(fp, dp, f_list); 1656 fp->f_count++; 1657 mutex_exit(&fp->f_lock); 1658 mutex_exit(&filelist_lock); 1659 1660 /* 1661 * Flush all data from the socket's receive buffer. 1662 * This will cause files referenced only by the 1663 * socket to be queued for close. 1664 */ 1665 so = fp->f_data; 1666 solock(so); 1667 sorflush(so); 1668 sounlock(so); 1669 1670 /* Re-lock and restart from where we left off. */ 1671 closef(fp); 1672 mutex_enter(&filelist_lock); 1673 np = LIST_NEXT(dp, f_list); 1674 LIST_REMOVE(dp, f_list); 1675 } 1676 } 1677 1678 /* 1679 * Garbage collector thread. While SCM_RIGHTS messages are in transit, 1680 * wake once per second to garbage collect. Run continually while we 1681 * have deferred closes to process. 1682 */ 1683 static void 1684 unp_thread(void *cookie) 1685 { 1686 file_t *dp; 1687 1688 /* Allocate a dummy file for our scans. */ 1689 if ((dp = fgetdummy()) == NULL) { 1690 panic("unp_thread"); 1691 } 1692 1693 mutex_enter(&filelist_lock); 1694 for (;;) { 1695 KASSERT(mutex_owned(&filelist_lock)); 1696 if (SLIST_EMPTY(&unp_thread_discard)) { 1697 if (unp_rights != 0) { 1698 (void)cv_timedwait(&unp_thread_cv, 1699 &filelist_lock, hz); 1700 } else { 1701 cv_wait(&unp_thread_cv, &filelist_lock); 1702 } 1703 } 1704 unp_gc(dp); 1705 } 1706 /* NOTREACHED */ 1707 } 1708 1709 /* 1710 * Kick the garbage collector into action if there is something for 1711 * it to process. 1712 */ 1713 static void 1714 unp_thread_kick(void) 1715 { 1716 1717 if (!SLIST_EMPTY(&unp_thread_discard) || unp_rights != 0) { 1718 mutex_enter(&filelist_lock); 1719 cv_signal(&unp_thread_cv); 1720 mutex_exit(&filelist_lock); 1721 } 1722 } 1723 1724 void 1725 unp_dispose(struct mbuf *m) 1726 { 1727 1728 if (m) 1729 unp_scan(m, unp_discard_later, 1); 1730 } 1731 1732 void 1733 unp_scan(struct mbuf *m0, void (*op)(file_t *), int discard) 1734 { 1735 struct mbuf *m; 1736 file_t **rp, *fp; 1737 struct cmsghdr *cm; 1738 int i, qfds; 1739 1740 while (m0) { 1741 for (m = m0; m; m = m->m_next) { 1742 if (m->m_type != MT_CONTROL || 1743 m->m_len < sizeof(*cm)) { 1744 continue; 1745 } 1746 cm = mtod(m, struct cmsghdr *); 1747 if (cm->cmsg_level != SOL_SOCKET || 1748 cm->cmsg_type != SCM_RIGHTS) 1749 continue; 1750 qfds = (cm->cmsg_len - CMSG_ALIGN(sizeof(*cm))) 1751 / sizeof(file_t *); 1752 rp = (file_t **)CMSG_DATA(cm); 1753 for (i = 0; i < qfds; i++) { 1754 fp = *rp; 1755 if (discard) { 1756 *rp = 0; 1757 } 1758 (*op)(fp); 1759 rp++; 1760 } 1761 } 1762 m0 = m0->m_nextpkt; 1763 } 1764 } 1765 1766 void 1767 unp_mark(file_t *fp) 1768 { 1769 1770 if (fp == NULL) 1771 return; 1772 1773 /* If we're already deferred, don't screw up the defer count */ 1774 mutex_enter(&fp->f_lock); 1775 if (fp->f_flag & (FMARK | FDEFER)) { 1776 mutex_exit(&fp->f_lock); 1777 return; 1778 } 1779 1780 /* 1781 * Minimize the number of deferrals... Sockets are the only type of 1782 * file which can hold references to another file, so just mark 1783 * other files, and defer unmarked sockets for the next pass. 1784 */ 1785 if (fp->f_type == DTYPE_SOCKET) { 1786 unp_defer++; 1787 KASSERT(fp->f_count != 0); 1788 atomic_or_uint(&fp->f_flag, FDEFER); 1789 } else { 1790 atomic_or_uint(&fp->f_flag, FMARK); 1791 } 1792 mutex_exit(&fp->f_lock); 1793 } 1794 1795 static void 1796 unp_discard_now(file_t *fp) 1797 { 1798 1799 if (fp == NULL) 1800 return; 1801 1802 KASSERT(fp->f_count > 0); 1803 KASSERT(fp->f_msgcount > 0); 1804 1805 mutex_enter(&fp->f_lock); 1806 fp->f_msgcount--; 1807 mutex_exit(&fp->f_lock); 1808 atomic_dec_uint(&unp_rights); 1809 (void)closef(fp); 1810 } 1811 1812 static void 1813 unp_discard_later(file_t *fp) 1814 { 1815 1816 if (fp == NULL) 1817 return; 1818 1819 KASSERT(fp->f_count > 0); 1820 KASSERT(fp->f_msgcount > 0); 1821 1822 mutex_enter(&filelist_lock); 1823 if (fp->f_unpcount++ == 0) { 1824 SLIST_INSERT_HEAD(&unp_thread_discard, fp, f_unplist); 1825 } 1826 mutex_exit(&filelist_lock); 1827 } 1828
Indexes created Sun Sep 21 20:09:37 GMT 2025